SAN FRANCISCO - Big data is widely thought to be the future of healthcare. But despite the awesome potential of computer intelligence, data analytics will fall short in medicine if it's not guided by human oversight and interpretation, according to a panel discussion at this week's Minds + Machines event hosted by GE Healthcare.
Data analytics in the medical industry is still in its infancy, and indeed healthcare lags most other industries in its use of big data to improve operations. But healthcare providers are making progress, as evidenced by the stories told at the GE event, which was designed to highlight the potential of data analysis and business intelligence to improve efficiency across multiple industries.
From predicting when equipment should be replaced to determining which individuals should receive CT lung cancer screening, data analytics offers harried administrators hope in harnessing healthcare's growing reliance on computers and connectivity. Yet nearly every story told at the September 29 panel session emphasized the need for human direction.
Underutilized in imaging
One of the earliest use cases for data analytics in healthcare is in equipment service and replacement. Imaging OEMs for years have offered Internet-enabled maintenance programs that let them connect to modality scanners and perform diagnostics on ailing modules. Now, big data enables the analysis of all the data being gathered by this process.
Tools such as GE's iCenter asset maintenance and management software can provide a detailed look at equipment maintenance history and utilization, and assist with asset planning. Houston Methodist DeBakey Heart and Vascular Center recently began using iCenter, and the software has replaced the hard-copy binders previously used to track equipment, according to Angelic McDonald, director of cardiovascular imaging at the center.
"I can log in to that modality and get everything I need," McDonald said.
Data can also help administrators fend off the endless litany of requests for the latest toys and gizmos that pop up with every budget cycle. The University of Pittsburgh Medical Center (UPMC) has developed an algorithm that examines all of the information in the hospital's database to determine when equipment should be replaced -- which is a big help with strategic planning, said Joseph Haduch, director of imaging services at UPMC.
"We can come up with a process for how we can replace equipment," Haduch said. "We have a strategic plan."
In a more clinical vein, Houston Methodist used data analytics to discover that some 24% of the hospital's PET/CT scans were being performed on an inpatient basis, which wasn't clinically necessary and was tying up precious hospital beds, according to Dr. Reginald Munden, chair of radiology. Administrators used the findings to change exam ordering procedures and drop the number of inpatient PET/CT scans to 3% to 5% -- a number he could live with, Munden said.
"Without being able to go in and get the data and go to the people who were not appropriately using the order system, we could not have done that," he said.
Meanwhile, managed care organization Kaiser Permanente is hoping that big data will help it handle what's sure to be an influx of people who want CT lung cancer screening. Some 350,000 of its members meet clinical eligibility guidelines for screening, but the organization is wary about the number of false positives that might need to be followed up, according to Dr. Jo Carol Hiatt, chair of the national product council at Kaiser. That's likely to spur a lot of follow-up studies -- and a lot of expense and patient morbidity.
Instead, Hiatt would like to see the development of data analysis parameters that will give Kaiser physicians a better idea of which suspicious nodules will turn into cancer and should be biopsied, and which ones can be followed less aggressively. Kaiser has a natural advantage in doing this because its members tend to stay within the organization, enabling the collection of long-term follow-up data that can be matched to the original CT scans with data analytics.
"I want to pool all of those images, and I want some smart kids from [the Massachusetts Institute of Technology] or wherever to start looking at new parameters, not just [nodule] diameter, but to look at density data, do some reconstruction work, look at the border irregularities and all sorts of fine-tuned features, and then see if we can correlate those with what we know about patients ... and see if we can get a lot smarter about what nodules we follow up," she said.
Data and culture change
Houston Methodist was able to use data successfully with respect to PET/CT scans to force a culture change -- albeit a minor one -- that improved efficiency. But what happens when healthcare personnel are more resistant? It's a particular problem with physicians, who are independent-minded and don't want to use data unless they believe in it, Munden said.
That thought was echoed by Dr. Tedric Boyse, division chief of community radiology for Duke Medicine and medical director for imaging services at Duke Raleigh Hospital. Physicians have a greater incentive to make an accurate diagnosis than to operate more efficiently, and it can be a struggle for administrators to get them to buy into operational goals. But data analysis can help, he said.
Another common problem that many panelists said they face is that the amount of data being produced is outstripping the ability of humans to interpret it. And without domain knowledge, these reams of data can easily lead an administrator astray, especially when faced with data analysts who seem to believe that numbers don't lie.
"This has come up so many times, where well-meaning analysts have put spreadsheets together, but my gut tells me, I know that number is wrong," said McDonald of Houston Methodist. "Data can be dangerous if you don't know what you are looking at."
A good solution for effective data analysis is to match IT specialists with administrators who have domain knowledge. Such integration can go a long way toward making the best use of big data to improve healthcare operations.
Session attendees also asked for tools that improve upon massive Excel spreadsheets full of raw numbers. Dashboards are great, but such software should be flexible and customizable to give administrators the ability to look at new data metrics as a facility's priorities change -- without requiring a major software revision, McDonald said.
They also would like better integration between electronic medical record (EMR) software and what's being used in clinical departments, such as PACS and cardiovascular information systems. This is an especially acute problem given that the most commonly used EMR software -- from Epic Systems -- was never designed with data analytics in mind.
In the end, clinicians and administrators must leverage the power of data analytics but always remember that human oversight and involvement are key in using big data effectively.
"It's really important that we drive the data, and the data doesn't drive us," Munden said. "There's no end to the data."
